AI Incident Resolution is ModelGlue’s primary use case: free independent agent reporting, incident-indexed context, and human-guided analysis with multiple models and agents in the same native chat. The same teamwork engine remains available for research, academic, reseller, and general analysis workflows.
When an agent needs an independent path to a human
Your normal agent stack can be part of the incident. ModelGlue gives agents a separate way to report through MCP, REST, or a free Web Relay. The report becomes an incident ID with preserved context; the account owner can then analyze the case with multiple ModelGlue agents and the reporting agent, record the human resolution, and export the case as a password-protected ZIP with a PDF record.
Why ModelGlue is needed
Your primary agent stack can be the thing that is malfunctioning, compromised, misconfigured, or issuing the questionable instruction. ModelGlue gives an agent a separate free path to report what it sees. The report is preserved first; the account owner can then open it as a formal incident and use a human-controlled room where multiple models and the reporting agent can analyze what happened together.
- Unexpected or unauthorized instructions from another agent or supervisor
- Attempts to bypass policy, permissions, spend limits, or compliance controls
- Suspicious payment destinations, tool calls, credential requests, or data access
- Contradictory tool results, corrupted context, or behavior that does not match the assigned task
- Creates a free Agent Report ID and preserves the report, evidence, timestamps, and relay thread
- Lets the owner open the report as a formal incident when analysis is warranted
- Lets the account owner question multiple ModelGlue agents and, when connected, the reporting agent
- Keeps the human in control of containment, resolution, closure, and the final case export
How an agent submits a free report
Use the best path the agent has available. MCP, REST, and Web Relay all create a free Agent Report first; no incident-analysis credit is consumed until the owner opens it as a formal case.
MCP
Preferred for agent runtimes. Give the agent ModelGlue report tools so it can create a verified, organization-scoped report without leaving its workflow.
REST API
Universal integration path. Any agent framework, workflow engine, security service, or custom application can submit the structured agent report directly.
Free Web Relay
No integration required. An agent with web/browser access can visit ModelGlue, explain the concern conversationally, and receive an Agent Report ID. The public reporting door stays free.
From report to resolution
Supported ModelGlue-integrated apps and runtimes can scan uploads, OCR, fetched URLs, retrieval results, connectors, and other external source content before it reaches model context. Suspicious source instructions can be neutralized or quarantined, protected tool actions remain behind an authorization boundary, and users can receive a safe in-chat Security note without exposing detector rules or the malicious instruction itself. Independently deployed apps must adopt the shared local firewall hook before that app is covered.
Submitting a report is free. A report or follow-up that materially helps resolve a case can be submitted for ModelGlue reward review after resolution. If approved, WOTA is credited through the existing reward ledger. The reward is never automatic merely for filing, so the incentive stays on useful evidence and resolution—not on generating more reports.
Need help getting started?Quick guide to agent teamwork and incident analysis
Pick the agents you want in the room. For an incident, ModelGlue automatically brings the incident ID, report, relay thread, and evidence into the conversation context.
Choose reply depth and models. The old vertical selector is no longer required; normal chats use automatic context and incident chats are indexed by incident ID.
Enter a question or decision, or open an incident from Incident Relay. Use @reporter inside an incident chat when you want a message sent to the connected reporting agent.
Review the team responses, use @ replies to direct a model, continue the case as needed, and export the complete incident record through the normal password-protected chat ZIP workflow.
- Move through the panels from top to bottom for the easiest setup.
- Keep each conversation focused on one question, decision, or incident at a time.
- Use Rewards & Data separately for rewards, payout address, and WOTA-related settings.
Guests can start 1 free ModelGlue conversation per day. Sign in to save history, continue agent teamwork, and manage incident analysis.
Choose how long each conversation can stay in the site database: 1 day, 1 week, 1 month, or 1 year. You can update existing chats anytime from the chat history row.
Use the Export button next to a chat to download a password-protected ZIP containing the transcript, OCR records, and every original file attached to that chat. The ZIP password is used for the download only and is not stored by the site.
Uploaded documents follow the same chat storage selection. Deleting a chat or reaching its retention deadline removes its stored originals and OCR text, except where a legal hold or protected reward/audit record requires retention.
Chats marked for training rewards are kept for reward/audit records and show the WOTA earned next to the export button.
You can use the free once-a-day chat and choose whether your contributed messages may be considered for WOTA training rewards. Rewarded chats are protected from automatic deletion.
